Chapter 2

Basic Theory of Cavity Optomechanics

Aashish A. Clerk and Florian Marquardt

Abstract This chapter provides a brief basic introduction to the theory used to
describe cavity-optomechanical systems. This can serve as background information
to understand the other chapters of the book. We first review the Hamiltonian and
show how it can be approximately brought into quadratic form. Then we discuss
the classical dynamics both in the linear regime (featuring optomechanical damping,
optical spring, strong coupling, and optomechanically induced transparency) and
in the nonlinear regime (optomechanical self-oscillations and attractor diagram).
Finally, we discuss the quantum theory of optomechanical cooling, using the powerful
and versatile quantum noise approach.

2.1 The Optomechanical Hamiltonian

Cavity optomechanical systems display a parametric coupling between the mechani-

cal displacement x̂ of a mechanical vibration mode and the energy stored inside a
radiation mode. That is, the frequency of the radiation mode depends on x̂ and can be
written in the form ωopt (x̂). When this dependence is Taylor-expanded, it is usually
sufficient to keep the linear term, and we obtain the basic cavity-optomechanical
Hamiltonian
 
Ĥ0 =  ωopt (0) − G x̂ â † â + ΩM b̂† b̂ + · · · (2.1)

A. A. Clerk (B)
Department of Physics, McGill University, Montreal, Canada
e-mail: clerk@physics.mcgill.ca
F. Marquardt
Institute for Theoretical Physics, Universität Erlangen-Nürnberg, Erlangen, Germany
e-mail: Florian.Marquardt@fau.de

M. Aspelmeyer et al. (eds.), Cavity Optomechanics, Quantum Science and Technology, 5

DOI: 10.1007/978-3-642-55312-7_2, © Springer-Verlag Berlin Heidelberg 2014
6 A. A. Clerk and F. Marquardt

Fig. 2.1 A typical system in

cavity optomechanics consists
of a laser-driven optical cav-
ity whose light field exerts
a radiation pressure force
on a vibrating mechanical optical mechanical
laser
resonator cavity mode

We have used ΩM to denote the mechanical frequency, â † â is the number of photons

circulating inside the optical cavity mode, and b̂† b̂ is the number of phonons inside
the mechanical mode of interest. Here G is the optomechanical frequency shift per
displacement, sometimes also called the “frequency pull parameter”, that charac-
terizes the particular system. For a simple Fabry–Perot cavity with an oscillating
end-mirror (illustrated in Fig. 2.1), one easily finds G = ωopt /L, where L is the
length of the cavity. This already indicates that smaller cavities yield larger coupling
strengths. A detailed derivation of this Hamiltonian for a model of a wave field inside
a cavity with a moving mirror can be found in [1]. However, the Hamiltonian is far
more general than this derivation (for a particular system) might suggest: Whenever
mechanical vibrations alter an optical cavity by leading to distortions of the boundary
conditions or changes of the refractive index, we expect a coupling of the type shown
here. The only important generalization involves the treatment of more than just a
single mechanical and optical mode (see the remarks below).
A coupling of the type shown here is called ‘dispersive’ (in contrast to a ‘dissi-
pative’ coupling, which would make κ depend on the displacement). Note that we
have left out the terms responsible for the laser driving and the decay (of photons
and of phonons), which will be dealt with separately in the following.
From this Hamiltonian, it follows that the radiation pressure force is

F̂rad = G â † â . (2.2)

After switching to a frame rotating at the incoming laser frequency ω L , we intro-

duce the detuning Δ = ω L − ωopt (0), such that we get

Ĥ = −Δâ † â − G x̂ â † â + ΩM b̂† b̂ + · · · (2.3)

It is now possible to write the displacement x̂ = xZPF (b̂ + b̂† ) in terms of the phonon
creation and annihilation operators, where xZPF = (/2m eff ΩM )1/2 is the size of the
2 Basic Theory of Cavity Optomechanics 7

Fig. 2.2 After linearization,

the standard system in cavity
optomechanics represents two
coupled harmonic oscillators,
one of them mechanical (at a
frequency ΩM ), the other opti-
cal (at an effective frequency mechanical oscillator driven optical cavity
given by the negative detuning
−Δ = ωopt (0) − ωL )
(decay rate ) (decay rate )

mechanical ground state wave function (“mechanical zero-point fluctuations”). This

leads to
 
Ĥ = −Δâ † â − g0 b̂ + b̂† â † â + ΩM b̂† b̂ + · · · . (2.4)

Here g0 = GxZPF represents the coupling between a single photon and a single
phonon. Usually g0 is a rather small frequency, much smaller than the cavity decay
rate κ or the mechanical frequency ΩM . However, the effective photon-phonon cou-
pling can be boosted by increasing the laser drive, at the expense of introducing a
coupling that is only quadratic (instead of cubic as in the original Hamiltonian). To
see this, we set â = α + δ â, where
  α is the average light field amplitude produced
by the laser drive (i.e. α = â in the absence of optomechanical coupling), and
δ â represents the small quantum fluctuations around that constant amplitude. If we
insert this into the Hamiltonian and only keep the terms that are linear in α, we obtain
  
Ĥ(lin) = −Δδ â † δ â − g b̂ + b̂† δ â + δ â † + ΩM b̂† b̂ + · · · (2.5)

This is the so-called “linearized” optomechanical Hamiltonian (where the equations

of motion for δ â and b̂ are in fact linear). Here g = g0 α is the enhanced, laser-
tunable optomechanical coupling strength, and for simplicity we have assumed α to
be real-valued (otherwise a simple unitary transformation acting on δ â can bring the
Hamiltonian to the present form, which is always possible unless two laser-drives
are involved). We have thus arrived at a rather simple system: two coupled harmonic
oscillators (Fig. 2.2).
Note that we have omitted the term −g0 |α|2 (b̂ + b̂† ), which represents a con-
stant radiation pressure force acting on the mechanical resonator and would lead to
a shift of the resonator’s equilibrium position. We can imagine (as is usually done in
these cases), that this shift has already been taken care of and x̂ is measured from the
new equilibrium position, or that this leads to a slightly changed “effective detun-
ing” Δ̄ (which will be the notation we use further below when solving the classical
equations of motion). In addition, we have neglected the term −g0 δ â † δ â(b̂ + b̂† ),
under the assumption that this term is “small”. The question when exactly this term
may start to matter and lead to observable consequences is a subject of ongoing
research (it seems that generally speaking g0 /κ > 1 is required).
8 A. A. Clerk and F. Marquardt

As will be explained below, almost all of the elementary properties of cavity-

optomechanical systems can be explained in terms of the linearized Hamiltonian.
Of course, the Hamiltonian in Eq. (2.1) represents an approximation (usually, an
extremely good one). In particular, we have omitted all the other mechanical normal
modes and all the other radiation modes. The justification for omitting the other
optical modes would be that only one mode is driven (nearly) resonantly by the laser.
With regard to the mechanical mode, optomechanical cooling or amplification in the
resolved-sideband regime (κ < M ) usually affects only one mode, again selected
by the laser frequency. Nevertheless, these simplistic arguments can fail, e.g. when
κ is larger than the spacing between mechanical modes, when the distance between
two optical modes matches a mechanical frequency, or when the dynamics becomes
nonlinear, with large amplitudes of mechanical oscillations.
Cases where the other modes become important display an even richer dynamics
than the one we are going to investigate below for the standard system (one mechan-
ical mode, one radiation mode). Interesting experimental examples for the case of
two optical modes and one mechanical mode can be found in the chapter by Bahl
and Carmon (on Brillouin optomechanics), and in the contribution by Jack Sankey
(on the membrane-in-the-middle setup).
In the following sections, we give a brief, self-contained overview of the most
important basic features of this system, both in the classical regime and in the quantum
regime. A more detailed introduction to the basics of the theory of cavity optome-
chanics can also be found in the recent review [2].

2.2 Classical Dynamics

The most important properties of optomechanical systems can be understood already

in the classical regime. As far as current experiments are concerned, the only signif-
icant exception would be the quantum limit to cooling, which will be treated further
below in the sections on the basics of quantum optomechanics.

2.2.1 Equations of Motion

In the classical regime, we assume both the mechanical oscillation amplitudes and the
optical amplitudes to be large, i.e. the system contains many photons and phonons.
As a matter of fact, much of what we will say is also valid in the regime of small
amplitudes, when only a few photons and phonons are present. This is because in
that regime the equations of motion can be linearized, and the expectation values of a
quantum system evolving according to linear Heisenberg equations of motion in fact
follow precisely the classical dynamics. The only aspect missing from the classical
description in the linearized regime is the proper treatment of the quantum Langevin
noise force, which is responsible for the quantum limit to cooling mentioned above.
2 Basic Theory of Cavity Optomechanics 9

We write down the classical equations for the position x(t) and for the complex
light field amplitude α(t) (normalized such that |α|2 would be the photon number in
the semiclassical regime):

ẍ = −ΩM 2
(x − x0 ) − ΓM ẋ + (Frad + Fext (t))/m eff (2.6)
κ
α̇ = [i(Δ + Gx) − κ/2]α + αmax (2.7)
2

Here Frad = G |α|2 is the radiation pressure force. The laser amplitude enters the
term αmax in the second equation, where we have chosen a notation such that α =
αmax on resonance (Δ = 0) in the absence of the optomechanical interaction (G = 0).
Note that the dependence on  in this equations vanishes once we express the photon
number in terms of the total light energy E stored inside the cavity: |α|2 = E /ωL .
This confirms that we are dealing with a completely classical problem, in which 
will not enter any end-results if they are expressed in terms of classical quantities like
cavity and laser frequency, cavity length, stored light energy (or laser input power),
cavity decay rate, mechanical decay rate, and mechanical frequency. Still, we keep the
present notation in order to facilitate later comparison with the quantum expressions.

2.2.2 Linear Response of an Optomechanical System

We have also added an external driving force Fext (t) to the equation of motion for x(t).
This is because our goal now will be to evaluate the linear response of the mechanical
system to this force. The idea is that the linear response will display a mechanical
resonance that turns out to be modified due to the interaction with the light field. It
will be shifted in frequency (“optical spring effect”) and its width will be changed
(“optomechanical damping or amplification”). These are the two most important
elementary effects of the optomechanical interaction. Optomechanical effects on
the damping rate and on the effective spring constant have been first analyzed and
observed (in a macroscopic microwave setup) by Braginsky and co-workers already
at the end of the 1960s [3].
First one has to find the static equilibrium position, by setting ẋ = 0 and α̇ = 0
and solving the resulting set of coupled nonlinear algebraic equations. If the light
intensity is large, there can be more than one stable solution. This ‘static bistability’
was already observed in the pioneering experiment on optomechanics with optical
forces by the Walther group in the 1980s [4]. We now assume that such a solution
has been found, and we linearize around it: x(t) = x̄ + δx(t) and α(t) = ᾱ + δα(t).
Then the equations for δx and δα read:

G  ∗ Fext (t)
δ ẍ(t) = −2M δx − ΓM δ ẋ + ᾱ δα + ᾱδα ∗ + (2.8)
m eff m eff
δ α̇(t) = [i Δ̄ − κ/2]δα + i G ᾱδx (2.9)
10 A. A. Clerk and F. Marquardt

Note that we have introduced the effective detuning Δ̄ = Δ + G x̄, shifted due to
the static mechanical displacement (this is often not made explicit in discussions
of optomechanical systems, although it can become important for larger displace-
ments). We are facing a linear set of equations, which in principle can be solved
straightforwardly by going to Fourier space and inverting a matrix. There is only one
slight difficulty involved here, which is that the equations also contain the complex
conjugate δα ∗ (t). If we were to enter with an ansatz δα(t)∝e−iωt , this automatically
generates terms ∝e+iωt at the negative frequency as well. In some cases, this may be
neglected (i.e. dropping the term δα ∗ from the equations), because the term δα ∗ (t)
is not resonant (this is completely equivalent to the “rotating wave approximation”
in the quantum treatment). However, here we want to display the full solution.

We nowiωt introduce the Fourier transform of any quantity A(t) in the form A[ω] ≡
dt A(t)e . Then, in calculating the response to a force given by Fext [ω], we have
to consider the fact that (δα ∗ ) [ω] = (δα[−ω])∗ . The equation for δα[ω] is easily
solved, yielding δα[ω] = χc (ω)i G ᾱδx[ω], with χc (ω) = [−iω − i Δ̄ + κ/2]−1 the
response function of the cavity. When we insert this into the equation for δx[ω],
we exploit (δα ∗ ) [ω] = (δα[−ω])∗ as well as (δx[−ω])∗ = δx[ω], since δx(t) is
real-valued. The result for the mechanical response is of the form

Fext [ω]
δx[ω] =  2  ≡ χx x (ω)Fext [ω]. (2.10)
m eff ΩM − ω2 − iωΓM + Σ(ω)

Here we have combined all the terms that depend on the optomechanical interaction
into the quantity Σ(ω) in the denominator. It is equal to

Σ(ω) = −iG 2 |ᾱ|2 χc (ω) − χc∗ (−ω) . (2.11)

Note that the prefactor can also be rewritten as G 2 |ᾱ|2 = 2m eff ΩM g 2 , by inserting
the expression for xZPF = (/2m eff ΩM )−1/2 and using (GxZPF |ᾱ|)2 = g 2 .
One may call Σ the “optomechanical self-energy” [5]. This is in analogy to the
self-energy of an electron appearing in the expression for its Green’s function, which
summarizes the effects of the interaction with the electron’s environment (photons,
phonons, other electrons, ...).
If the coupling is weak, the mechanical linear response will still have a single
resonance, whose properties are just modified by the presence of the optomechanical
interaction. In that case, close inspection of the denominator in Eq. (2.10) reveals
the meaning of both the imaginary and the real part of Σ, which we evaluate at the
original resonance frequency ω = Ω. The imaginary part describes some additional
optomechanical damping, induced by the light field:

1
Γopt = − ImΣ(Ω)
m eff ΩM
 
1 1
= g2 κ  2  2 −  2  2 (2.12)
ΩM + Δ̄ + κ2 ΩM − Δ̄ + κ2
2 Basic Theory of Cavity Optomechanics 11

Fig. 2.3 Optomechanical damping rate (left) and frequency shift (right), as a function of the effec-
tive detuning Δ̄. The different curves depict the results for varying cavity decay rate, running in the
interval κ/ΩM = 0.2, 0.4, . . . , 5 (the largest values are shown as black lines). We keep the intra-
cavity energy fixed (i.e. g is fixed). Note that the damping rate Γopt has been rescaled by g 2 /κ, which
represents the parametric dependence of Γopt in the resolved-sideband regime κ < ΩM . In addition,
note that we chose to plot the frequency shift in terms of δ(Ω 2 ) ≈ 2ΩM δΩ (for small δΩ  ΩM )

The real part describes a shift of the mechanical frequency (“optical spring”):

1
δ(Ω 2 ) = ReΣ(Ω)
m eff
 
ΩM + Δ̄ ΩM − Δ̄
= 2ΩM g 2
 2  κ 2 −  2  2 . (2.13)
ΩM + Δ̄ + 2 ΩM − Δ̄ + κ2

Both of these are displayed in Fig. 2.3. They are the results of “dynamical backac-
tion”, where the (possibly retarded) response of the cavity to the mechanical motion
acts back on this motion.

2.2.3 Strong Coupling Regime

When the optomechanical coupling rate g becomes comparable to the cavity damping
rate κ, the system enters the strong coupling regime. The hallmark of this regime
is the appearance (for red detuning) of a clearly resolved double-peak structure in
the mechanical (or optical) susceptibility. This peak splitting in the strong coupling
regime was first predicted in [5], then analyzed further in [6] and finally observed
experimentally for the first time in [7]. This comes about because the mechanical
resonance and the (driven) cavity resonance hybridize, like any two coupled harmonic
oscillators, with a splitting 2g set by the coupling. In order to describe this correctly,
we have to retain the full structure of the mechanical susceptibility, Eq. (2.10) at all
12 A. A. Clerk and F. Marquardt

Fig. 2.4 Optomechanical strong coupling regime, illustrated in terms of the mechanical suscepti-
bility. The figures show the imaginary part of χx x (ω) = 1/(m(Ω 2 − ω2 − iωΓ ) + Σ(ω)). Left
Imχx x (ω) as a function of varying coupling strength g, set by the laser drive, for red detuning on
resonance, Δ̄ = −Ω. A clear splitting develops around g/κ = 0.5. Right Imχx x (ω) as a function
of varying detuning Δ̄ between the laser drive and the cavity resonance, for fixed g/κ = 0.5

frequencies, without applying the previous approximation of evaluating Σ(ω) in the

vicinity of the resonance (Fig. 2.4).

2.2.4 Optomechanically Induced Transparency

We now turn to the cavity response to a weak additional probe beam, which can
be treated in analogy to the mechanical response discussed above. However, an
interesting new feature develops, due to the fact that usually Γ  κ. Even for
g  κ, the cavity response shows a spectrally sharp feature due to the optomechanical
interaction, and its width is given by Γ = ΓM + Γopt . This phenomenon is called
“optomechanically induced transparency” [8, 9].
We can obtain the modified cavity response by imagining that there is no mechan-
ical force (Fext = 0), but instead there is an additional weak laser drive, which enters
as · · · + δα L e−iωt on the right-hand-side of Eq. (2.9). By solving the coupled set of
equations, we arrive at a modified cavity response

δα(t) = χceff (ω)δα L e−iωt , (2.14)

where we find

χceff (ω) = χc (ω) · 1 + 2im eff ΩM g 2 χc (ω)χx x (ω) . (2.15)
2 Basic Theory of Cavity Optomechanics 13

Fig. 2.5 Optomechanically induced transparency: Modification of the cavity response due to the
interaction with the mechanical degree of freedom. We show Reχceff (ω) as a function of the detuning
ω between the weak probe beam and the strong(er) control beam, for variable coupling g of the
control beam (left) and for variable detuning Δ̄ of the control beam versus the cavity resonance
(right, at g/κ = 0.1). We have chosen κ/ΩM = 0.2. Note that in the left plot, for further increases
in g, the curves shown here would smoothly evolve into the double-peak structure characteristic of
the strong-coupling regime

Note that in the present section ω has the physical meaning of the detuning between
the weak additional probe laser and the original (possibly strong) control beam at ω L .
That is: ω = ωprobe − ω L . The result is shown in Fig. 2.5. The sharp dip goes down to
zero when g 2 /(κΓM )  1. Ultimately, this result is an example of a very generic phe-
nomenon: If two oscillators are coupled and they have very different damping rates,
then driving the strongly damped oscillator (here: the cavity) can indirectly drive the
weakly damped oscillator (here: the mechanics), leading to a sharp spectral feature
on top of a broad resonance. In the context of atomic physics with three-level atoms,
this has been observed as “electromagnetically induced transparency”, and thus the
feature discussed here came to be called “optomechanically induced transparency”.
We note that for a blue-detuned control beam, the dip turns into a peak, signalling
optomechanical amplification of incoming weak radiation.

2.2.5 Nonlinear Dynamics

On the blue-detuned side (Δ̄ > 0), where Γopt is negative, the overall damping rate
ΓM + Γopt diminishes upon increasing the laser intensity, until it finally becomes
negative. Then the system becomes unstable and any small initial perturbation (e.g.
thermal fluctuations) will increase exponentially at first, until the system settles into
self-induced mechanical oscillations of a fixed amplitude: x(t) = x̄ + A cos(ΩM t).
This is the optomechanical dynamical instability (parametric instability), which
14 A. A. Clerk and F. Marquardt

amplitude laser
power

phase

bifurcation

Fig. 2.6 When increasing a control parameter, such as the laser power, an optomechanical system
can become unstable and settle into periodic mechanical oscillations. These correspond to a limit
cycle in phase space of some amplitude A, as depicted here. The transition is called a Hopf bifurcation

has been explored both theoretically [10, 11] and observed experimentally in var-
ious settings (e.g. [12–14] for radiation-pressure driven setups and [15, 16] for
photothermal light forces) (Fig. 2.6).
In order to understand the saturation of the amplitude A at a fixed finite value,
we have to take into account that the mechanical ocillation changes the pattern of
the light amplitude’s evolution. In turn, the overall effective damping rate, as aver-
aged over an oscillation period, changes as well. To capture this, we now introduce an
amplitude-dependend optomechanical damping rate. This can be done by noting that
a fixed damping rate Γ would give rise to a power loss (m eff Γ ẋ) ẋ = Γ m eff A2 /2.
Thus, we define
2
Γopt (A) ≡ − Frad (t)ẋ(t) . (2.16)
m eff A2

This definition reproduces the damping rate Γopt calculated above in the limit A → 0.
The condition for the value of the amplitude on the limit cycle is then simply given by

Γopt (A) + ΓM = 0. (2.17)

The result for Eq. (2.16) can be expressed in terms of the exact analytical solution
for the light field amplitude α(t) given the mechanical
 oscillations
 at amplitude
 2α L  
A. This solution is a Fourier series, |α(t)| =  ΩM n α̃n e inΩ M t
, involving Bessel
function coefficients. In the end, we find
 2  
κ g2 GA
Γopt (A) = 4 f , (2.18)
ΩM ΩM ΩM

where
1 ∗
f (a) = − Imα̃n+1 α̃n (2.19)
a n

with
1 Jn (−a)
α̃n = . (2.20)
2 in + κ/(2ΩM ) − i Δ̄/ΩM
2 Basic Theory of Cavity Optomechanics 15

Fig. 2.7 Optomechanical Attractor Diagram: The effective amplitude-dependent optomechanical

damping rate Γopt (A), as a function of the oscillation amplitude A and the effective detuning
Δ̄, for three different sidebands ratios κ/ΩM = 0.2, 1, 2, from left to right [Γopt in units of
γ0 ≡ 4 (κ/ΩM )2 g 2 /ΩM , blue positive/cooling; red negative/amplification]. The optomechanical
attractor diagram of self-induced oscillations is determined via the condition Γopt (A) = −ΓM .
The attractors are shown for three different values of the incoming laser power (as parametrized by
the enhanced optomechanical coupling g at resonance), with ΓM /γ0 = 0.1, 10−2 , 10−3 (white,
yellow, red)

Note that g denotes the enhanced optomechanical coupling at resonance (i.e. for
Δ̄ = 0), i.e. it characterizes the laser amplitude. Also note that Δ̄ includes an
amplitude-dependent shift due to a displacement of the mean oscillator position
x̄ by the radiation pressure force Frad . This has to be found self-consistently.
The resulting attractor diagram is shown in Fig. 2.7. It shows the possible limit
cycle amplitude(s) as a function of effective detuning Δ̄, such that the self-consistent
evaluation of x̄ has been avoided.
The self-induced mechanical oscillations in an optomechanical system are anal-
ogous to the behaviour of a laser above threshold. In the optomechanical case, the
energy provided by the incoming laser beam is converted, via the interaction, into
coherent mechanical oscillations. While the amplitude of these oscillations is fixed,
the phase depends on random initial conditions and may diffuse due to noise (e.g. ther-
mal mechanical noise or shot noise from the laser). Interesting features may therefore
arise when several such optomechanical oscillators are coupled, either mechanically
or optically. In that case, they may synchronize if the coupling is strong enough.
Optomechanical synchronization has been predicted theoretically [17, 18] and then
observed experimentally [19, 20]. At high driving powers, we note that the dynamics
is no longer a simple limit cycle but may instead become chaotic [21].

2.3 Quantum Theory

In the previous section, we have seen how a semiclassical description of the canon-
ical optomechanical cavity gives a simple, intuitive picture of optical spring and
optical damping effects. The average cavity photon number n̄ cav acts as a force on
16 A. A. Clerk and F. Marquardt

the mechanical resonator; this force depends on the mechanical position x, as changes
in x change the cavity frequency and hence the effective detuning of the cavity drive
laser. If n̄ cav were able to respond instantaneously to changes in x, we would only
have an optical spring effect; however, the fact that n̄ cav responds to changes in x
with a non-zero delay time implies that we also get an effective damping force from
the cavity.
In this section, we go beyond the semiclassical description and develop the full
quantum theory of our driven optomechanical system [5, 22, 23]. We will see that
the semiclassical expressions derived above, while qualitatively useful, are not in
general quantitatively correct. In addition, the quantum theory captures an important
effect missed in the semiclassical description, namely the effective heating of the
mechanical resonator arising from the fluctuations of the cavity photon number about
its mean value. These fluctuations play a crucial role, in that they set a limit to the
lowest possible temperature one can achieve via cavity cooling.

2.3.1 Basics of the Quantum Noise Approach to Cavity Backaction

We will focus here on the so-called “quantum noise” approach, where for a weak
optomechanical coupling, one can understand the effects of the cavity backaction
completely from the quantum noise spectral density of the radiation pressure force
operator (Fig. 2.8). This spectral density is defined as:

∞  
SFF [ω] = dteiωt F̂(t) F̂(0) (2.21)
−∞

where the average is taken over the state of the cavity at zero optomechanical cou-
pling, and   
F̂(t) ≡ G â † â − â † â (2.22)

is the noise part of the cavity’s backaction force operator (in the Heisenberg picture).
We start by considering the quantum origin of optomechanical damping, treating
the optomechanical interaction term in the Hamiltonian of Eq. (2.3) using per-
turbation theory. Via the optomechanical interaction, the cavity will cause transi-
tions between energy eigenstates of the mechanical oscillator, either upwards or
downwards in energy. Working to lowest order in the optomechanical coupling G,
these rates are described by Fermi’s Golden rule. A straightforward calculation (see
Sect. II B of Ref. [24]) shows that the Fermi’s Golden rule rate Γn,+ (Γn,− ) for a
transition taking the oscillator from n → n+1 quanta (n → n−1 quanta) is given by:
  2
1 1 xZPF
Γn,± = n+ ± SFF [∓ΩM ] (2.23)
2 2 2
2 Basic Theory of Cavity Optomechanics 17

Fig. 2.8 The noise spectrum

of the radiation pressure force
in a driven optical cavity.
This is a Lorentzian, peaked
at the (negative) effective
detuning. The transition rates
are proportional to the value
of the spectrum at +ΩM
(emission of energy into the
cavity bath) and at −ΩM
(absorption of energy by the
mechanical resonator)

-3 -2 -1 0 1 2 3

The optomechanical damping rate simply corresponds to the decay rate of the average
oscillator energy due to these transitions. One finds (see Appendix B of Ref. [24]):
   
1 1
Γopt = Γn,− − Γn,+ (2.24)
n (n + 1)
2
xZPF
= (SFF [ΩM ] − SFF [−ΩM ]) (2.25)
2
Note that one obtains simple linear damping (the damping is independent of the
amplitude of the oscillator’s motion). Also note that our derivation has neglected
the effects of the oscillator’s intrinsic damping ΓM , and thus is only valid if ΓM is
sufficiently small; we comment more on this at the end of the section.
There is a second way to derive Eq. (2.25) which is slightly more general, and
which allows us to calculate the optical spring constant kopt ; it also more closely
matches the heuristic reasoning that led to the semiclassical expressions of the pre-
vious section. We start from the basic fact that both Γopt and δΩM,opt arise from the
dependence of the average backaction force Frad on the mechanical position x. We
can calculate this dependence to lowest order in G using the standard equations of
quantum linear response (i.e. the Kubo formula):

∞
   
δ Frad (t) = − dt λFF t − t x̂ t (2.26)
−∞

where the causal force-force susceptibility λFF (t) is given by:

i  
λFF (t) = − θ (t) F̂(t), F̂(0) (2.27)

18 A. A. Clerk and F. Marquardt

Next, assume that the oscillator is oscillating, and thus x̂(t) = x0 cos ΩM t. We
then have:

δ Frad (t) = (−Re λFF [ΩM ] · x0 cos ΩM t) − (Im λFF [ΩM ] · x0 sin ΩM t) (2.28)
 
  d x̂
= −Δkopt x̂(t) − m eff Γopt (t) (2.29)
dt

Comparing the two lines above, we see immediately that the real and imaginary parts
of the Fourier-transformed susceptibility λFF [ΩM ] are respectively proportional to
the optical spring kopt and the optomechanical damping Γopt . The susceptibility
can in turn be related to SFF [ω]. In the case of the imaginary part of λFF [ω], a
straightforward calculation yields:
SFF [ω] − SFF [−ω]
−Im λFF [ω] = (2.30)
2
As a result, the definition of Γopt emerging from Eq. (2.29) is identical to that in
Eq. (2.25). The real part of λFF [ω] can also be related to SFF [ω] using a standard
k
Kramers-Kronig identity. Defining δΩM,opt ≡ 2m effoptΩM , one finds:

∞  
x2 dω 1 1
δΩM,opt = ZPF SFF [ω] − (2.31)
2 2π ΩM − ω ΩM + ω
−∞

Thus, a knowledge of the quantum noise spectral density SFF [ω] allows one to imme-
diately extract both the optical spring coefficient, as well as the optical damping rate.
We now turn to the effects of the fluctuations in the radiation pressure force, and the
effective temperature Trad which characterizes them. This too can be directly related
to SFF [ω]. Perhaps the most elegant manner to derive this is to perturbatively integrate
out the dynamics of the cavity [24, 25]; this approach also has the benefit of going
beyond simplest lowest-order-perturbation theory. One finds that the mechanical
resonator is described by a classical Langevin equation of the form:

m ẍ(t) = −(k + kopt )x(t) − mΓopt ẋ(t) + ξrad (t). (2.32)

The optomechanical damping Γopt and optical spring kopt are given respectively
by Eqs. (2.25) and (2.31), except that one should make the replacement ΩM →
ΩM ≡ Ω + δΩ
M M,opt in these equations. The last term ξrad (t) above represents the
fluctuating backaction force associated with photon number fluctuations in the cavity.
Within our approximations of weak optomechanical coupling and weak intrinsic
mechanical damping, this random force is Gaussian white noise, and is fully described
by the spectral density:
 
Sξrad ξrad [ω] = mΓopt coth ΩM /2k B Trad = mΓopt (1 + 2n̄ rad ) . (2.33)
2 Basic Theory of Cavity Optomechanics 19

Here, Trad is the effective temperature of the cavity backaction, and n̄ rad is the cor-
responding number of thermal oscillator quanta. These quantities are determined by
SFF [ω] via:
 

SFF ΩM + SFF −ΩM
1 + 2n̄ rad ≡  
(2.34)
SFF ΩM − SFF −ΩM

Note that as the driven cavity is not in thermal equilibrium, Trad will in general
depend on the value of ΩM ; a more detailed discussion of the concept of an effective
temperature is given in Ref. [24].
Turning to the stationary state of the oscillator, we note that Eq. (2.32) is identical
to the Langevin equation for an oscillator coupled to a thermal equilibrium bath at
temperature Trad . It thus follows that the stationary state of the oscillator will be
a thermal equilibrium state at a temperature Trad , and with an average number of
quanta n̄ rad . As far as the oscillator is concerned, Trad is indistinguishable from a
true thermodynamic bath temperature, even though the driven cavity is not itself in
thermal equilibrium.
The more realistic case is of course where we include the intrinsic damping and
heating of the mechanical resonator; even here, a similar picture holds. The intrinsic
dissipation can be simply accounted for by adding to the RHS of Eq. (2.32) a damping
term describing the intrinsic damping (rate. ΓM ), as well as a stochastic force term
corresponding to the fridge temperature T . The resulting Langevin equation still
continues to have the form of an oscillator coupled to a single equilibrium bath, where
the total damping rate due to the bath is ΓM + Γopt , and the effective temperature
Teff of the bath is determined by:

ΓM n 0 + Γopt n̄ rad
n eff = (2.35)
ΓM + Γopt

where n 0 is the Bose-Einstein factor corresponding to the bath temperature T :

1
n0 =  /k T
 (2.36)
exp ΩM B −1

We thus see that in the limit where Γopt  ΓM , the effective mechanical temperature
tends to the backaction temperature Trad . This will be the lowest temperature possible
via cavity cooling . Note that similar results may be obtained by using the Golden rule
transition rates in Eq. (2.23) to formulate a master equation describing the probability
pn (t) that the oscillator has n quanta at time t (see Sect. II B of Ref. [24]).
Before proceeding, it is worth emphasizing that the above results all rely on the
total mechanical bandwidth ΓM + Γopt being sufficiently small that one can ignore
the variance of SFF [ω] across the mechanical resonance. When this condition is not
satisfied, one can still describe backaction effects using the quantum noise approach,
20 A. A. Clerk and F. Marquardt

with a Langevin equation similar to Eq. (2.32). However, one now must include the
variation of SFF [ω] with frequency; the result is that the optomechanical damping
will not be purely local in time, and the stochastic part of the backaction force will
not be white.

2.3.2 Application to the Standard Cavity Optomechanical Setup

The quantum noise approach to backaction is easily applied to the standard opto-
mechanical cavity setup, where the backaction force operator F̂ is proportional to
the cavity photon number operator. To calculate its quantum noise spectrum in the
absence of any optomechanical coupling, we first write the equation of motion for the
cavity annihilation operator â in the Heisenberg picture, using standard input–output
theory [26, 27]:

d   √
â = −iωopt − κ/2 â − κ âin . (2.37)
dt

Here, âin describes the amplitude of drive laser, and can be decomposed as:
 
âin = e−iω L t āin + d̂in , (2.38)

where āin represents the classical amplitude of the drive laser (the input power is
given by Pin = ωopt |āin |2 ), and d̂in describes fluctuations in the laser drive. We
consider the ideal case where these are vacuum noise, i.e. there is only shot noise
in the incident laser drive, and no additional thermal or phase fluctuations. One thus
finds that d̂in describes operator white noise:
   
† †
d̂in (t)d̂in (t ) = [d̂in (t), d̂in (t )] = δ(t − t ) (2.39)

It is also useful to separate the cavity field operator into an average “classical” part
and a quantum part,
 
â = e−iω L t eiφ n̄ cav + d̂ (2.40)

√
where eiφ n̄ cav is the classical amplitude of the cavity field, and d̂ describes its
fluctuations.
It is now straightforward to solve Eq. (2.37) for d̂ in terms of d̂in . As we will be
interested in regimes where n̄ cav  1, we can focus on the leading-order-in-n̄ cav
term in the backaction force operator F:
  
F̂  G n̄ cav d̂ + d̂ † (2.41)
2 Basic Theory of Cavity Optomechanics 21

Using this leading-order expression along with the solution for d̂in and Eq. (2.39),
we find that the quantum noise spectral density SFF [ω] (as defined in Eq. (2.21)) is
given by:
κ
SFF [ω] = 2 G 2 n̄ cav (2.42)
(ω + Δ)2+ (κ/2)2

SFF [ω] is a simple Lorentzian, reflecting the cavity’s density of states, and is centred
at ω = −Δ, precisely the energy required to bring a drive photon onto resonance.
The form of SFF [ω] describes the final density of states for a Raman process where
an incident drive photon gains (ω > 0, anti-Stokes) or loses (ω < 0, Stokes) a
quanta |ω| of energy before attempting to enter the cavity. From Eq. (2.25), we
can immediately obtain an expression for the optomechanical damping rate; it will
be large if can make the density states associated with the anti-Stokes process at
frequency ΩM much larger than that of the Stokes process at the same frequency.

The optical spring coefficient also follows from Eq. (2.31).

We finally turn to n̄ rad , the effective temperature of the backaction (expressed as
a number of oscillator quanta). Using Eqs. (2.34) and (2.42), we find:

+ Δ)2 + (κ/2)2
(ΩM
n̄ rad = − Δ (2.43)
4ΩM

As discussed, n̄ rad represents the lowest possible temperature we can cool our
mechanical resonator to. As a function of drive detuning Δ, n̄ rad achieves a min-
imum value of
  2
 κ 2
n̄ rad  =  (2.44)
min 4ΩM 1 + 1 + (κ 2 /4Ω 2 )
M

for an optimal detuning of


2 + κ 2 /4.
Δ = − ΩM (2.45)

We thus see that if one is in the so-called good cavity limit ΩM  κ, and if the
detuning is optimized, one can potentially cool the mechanical resonator close to
its ground state. In this limit, the anti-Stokes process is on-resonance, while the
Stokes process is far off-resonance and hence greatly suppressed. The fact that the
effective temperature is small but non-zero in this limit reflects the small but non-zero
probability for the Stokes process, due to the Lorentzian tail of the cavity density of
states. In the opposite, “bad cavity” limit where ΩM  κ, we see that the minimum
of n̄ rad tends to κ/ΩM  1, while the optimal detuning tends to κ/2 (as anticipated
in the semiclassical approach).
Note that the above results are easily extended to the case where the cavity is
driven by thermal noise corresponding to a thermal number of cavity photons n cav,T .
For a drive detuning of Δ = −ΩM (which is optimal in the good cavity limit), one
22 A. A. Clerk and F. Marquardt

now finds that that the n̄ rad is given by [6]:

 2   2 
κ κ
n̄ rad = + n cav,T 1+2 (2.46)
4ΩM 4ΩM

As expected, one cannot backaction-cool a mechanical resonator to a temperature

lower than that of the cavity.

2.3.3 Results for a Dissipative Optomechanical Coupling

A key advantage of the quantum noise approach is that it can be easily applied to
alternate forms of optomechanical coupling. For example, it is possible have systems
where the mechanical resonator modulates both the cavity frequency as well as the
damping rate κ of the optical cavity [28, 29]. The position of the mechanical resonator
will now couple to both the cavity photon number (as in the standard setup), as well as
to the “photon tunnelling” term which describes the coupling of the cavity mode to the
extra-cavity modes that damp and drive it. Because of these two couplings, the form of
the effective backaction force operator F̂ is now modified from the standard setup.
Nonetheless, one can still go ahead and calculate the optomechanical backaction
using the quantum noise approach. In the simple case where the cavity is overcoupled
(and hence its κ is due entirely to the coupling to the port used to drive it), one finds
that the cavity’s backaction quantum force noise spectrum is given by [30, 31]:
2
  ω + 2Δ − 2G
κ
G 2κ n̄ cav Gκ
SFF [ω] = (2.47)
4κ (ω + Δ)2 + κ 2 /4

Here, G = −dωopt /d x is the standard optomechanical coupling, while G κ = dκ/d x

represents the dissipative optomechanical coupling. For G κ = 0, we recover the
Lorentzian spectrum of the standard optomechanical setup given in Eq. (2.42)
whereas for G κ = 0, SFF [ω] has the general form of a Fano resonance. Such
lineshapes arise as the result of interference between resonant and non-resonant
processes; here, the resonant channel corresponds to fluctuations in the cavity ampli-
tude, whereas the non-resonant channel corresponds to the incident shot noise fluc-
tuations on the cavity. These fluctuations can interfere destructively, resulting in
SFF [ω] = 0 at the special frequency ω = −2Δ + 2G/G κ . If one tunes Δ such that
this frequency coincides with −ΩM , it follows immediately from Eq. (2.34) that the
cavity backaction has an effective temperature of zero, and can be used to cool the
mechanical resonator to its ground state. This special detuning causes the destructive
interference to completely suppress the probability of the cavity backaction exciting
the mechanical resonator, whereas the opposite process of absorption is not sup-
pressed. This “interference cooling” does not require one to be in the good cavity
limit, and thus could be potentially useful for the cooling of low-frequency (relative
2 Basic Theory of Cavity Optomechanics 23

to κ) mechanical modes. However, the presence of internal loss in the cavity places
limits on this technique, as it suppresses the perfect destructive interference between
resonant and non-resonant fluctuations [30, 31].

References

1. C.K. Law, Phys. Rev. A 51, 2537 (1995)

2. M. Aspelmeyer, T.J. Kippenberg, F. Marquardt (2013), arXiv:1303.0733
3. V.B. Braginsky, A.B. Manukin, Sov. Phys. JETP 25, 653 (1967)
4. A. Dorsel, J.D. McCullen, P. Meystre, E. Vignes, H. Walther, Phys. Rev. Lett. 51, 1550 (1983)
5. F. Marquardt, J.P. Chen, A.A. Clerk, S.M. Girvin, Phys. Rev. Lett. 99, 093902 (2007)
6. J. Dobrindt, I. Wilson-Rae, T.J. Kippenberg, Phys. Rev. Lett. 101(26), 263602 (2008)
7. S. Groblacher, K. Hammerer, M.R. Vanner, M. Aspelmeyer, Nature 460, 724 (2009)
8. G.S. Agarwal, S. Huang, Phys. Rev. A 81, 041803 (2010)
9. S. Weis, R. Rivière, S. Deléglise, E. Gavartin, O. Arcizet, A. Schliesser, T.J. Kippenberg,
Science 330, 1520 (2010)
10. F. Marquardt, J.G.E. Harris, S.M. Girvin, Phys. Rev. Lett. 96, 103901 (2006)
11. M. Ludwig, B. Kubala, F. Marquardt, New J. Phys. 10, 095013 (2008)
12. H. Rokhsari, T.J. Kippenberg, T. Carmon, K. Vahala, Opt. Express 13, 5293 (2005)
13. T. Carmon, H. Rokhsari, L. Yang, T.J. Kippenberg, K.J. Vahala, Phys. Rev. Lett. 94, 223902
(2005)
14. T.J. Kippenberg, H. Rokhsari, T. Carmon, A. Scherer, K.J. Vahala, Phys. Rev. Lett. 95, 033901
(2005)
15. C. Höhberger, K. Karrai, in Nanotechnology 2004, Proceedings of the 4th IEEE conference on
nanotechnology (2004), p. 419
16. C. Metzger, M. Ludwig, C. Neuenhahn, A. Ortlieb, I. Favero, K. Karrai, F. Marquardt, Phys.
Rev. Lett. 101, 133903 (2008)
17. G. Heinrich, M. Ludwig, J. Qian, B. Kubala, F. Marquardt, Phys. Rev. Lett. 107, 043603 (2011)
18. C.A. Holmes, C.P. Meaney, G.J. Milburn, Phys. Rev. E 85, 066203 (2012)
19. M. Zhang, G. Wiederhecker, S. Manipatruni, A. Barnard, P.L. McEuen, M. Lipson, Phys. Rev.
Lett. 109, 233906 (2012)
20. M. Bagheri, M. Poot, L. Fan, F. Marquardt, H.X. Tang, Phys. Rev. Lett. 111, 213902 (2013)
21. T. Carmon, M.C. Cross, K.J. Vahala, Phys. Rev. Lett. 98, 167203 (2007)
22. I. Wilson-Rae, N. Nooshi, W. Zwerger, T.J. Kippenberg, Phys. Rev. Lett. 99, 093901 (2007)
23. C. Genes, D. Vitali, P. Tombesi, S. Gigan, M. Aspelmeyer, Phys. Rev. A 77, 033804 (2008)
24. A.A. Clerk, M.H. Devoret, S.M. Girvin, F. Marquardt, R.J. Schoelkopf, Rev. Mod. Phys. 82,
1155 (2010)
25. J. Schwinger, J. Math. Phys. 2, 407 (1961)
26. C.W. Gardiner, M.J. Collett, Phys. Rev. A 31(6), 3761 (1985)
27. C.W. Gardiner, P. Zoller, Quant. Noise (Springer, Berlin, 2000)
28. M. Li, W.H.P. Pernice, H.X. Tang, Phys. Rev. Lett. 103(22), 223901 (2009)
29. J.C. Sankey, C. Yang, B.M. Zwickl, A.M. Jayich, J.G.E. Harris, Nat. Phys. 6, 707 (2010)
30. F. Elste, S.M. Girvin, A.A. Clerk, Phys. Rev. Lett. 102, 207209 (2009)
31. F. Elste, A.A. Clerk, S.M. Girvin, Phys. Rev. Lett. 103, 149902(E) (2009)
http://www.springer.com/978-3-642-55311-0